Internal structure evidence of validity - GitHub Pages · Internal structure evidence of validity Dr Wan Nor Arifin Unit of Biostatistics and Research Methodology, Universiti Sains

Internal structure evidence of validity

Dr Wan Nor Arifin

Unit of Biostatistics and Research Methodology,Universiti Sains Malaysia.

[email protected] / wnarifin.github.io

Wan Nor Arifin, 2017. Internal structure evidence of validity by Wan Nor Arifin is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/.

10/10/2017 Q Validation Workshop 2017 2 / 27

Outlines

1.Measurement validity & reliability2.Classical validity3.The validity4.Factor analysis5.Reliability

10/10/2017 Q Validation Workshop 2017 3 / 27

1. Measurement validity & Reliability

• Measurement → Process of observing & recording.• Measurement validity → Accuracy.• Measurement reliability → Precision, consistency,

repeatability.

10/10/2017 Q Validation Workshop 2017 4 / 27

2. Classical validity

• 3Cs:1.Content2.Criterion3.Construct

10/10/2017 Q Validation Workshop 2017 5 / 27

3. The validity

• Unitary concept.• Degree of evidence → Purpose & Intended use of a

tool.• Evidence from 5 sources:

1.Content.2.Internal structure.3.Relations to other variables4.Response process.5.Consequences.

10/10/2017 Q Validation Workshop 2017 6 / 27

The validity

• Construct – Concept to be measured by a tool.• Construct = Concept = Domain = Idea• Internal structure evidence – How relationship between items & components

reflect construct.– Analyses:

1.Factor analysis2.Reliability

10/10/2017 Q Validation Workshop 2017 7 / 27

4. Factor Analysis

• Factoring• Factor analysis

10/10/2017 Q Validation Workshop 2017 8 / 27

Factoring

• Group things that have common concept.• Simplify.• Factoring = Grouping = Simplify• Factor → Construct

10/10/2017 Q Validation Workshop 2017 9 / 27

Intuitive factoring

Orange, motorcycle, bus, durian, banana, car

Anything in common?

10/10/2017 Q Validation Workshop 2017 10 / 27

Intuitive factoring

• Group them

Orange, durian, banana

Motorcycle, bus, car

into 2 groups

10/10/2017 Q Validation Workshop 2017 11 / 27

Intuitive factoring

• Name the groups

Fruit Motor vehicle

OrangeDurianBanana

MotorcycleBusCar

factor out the common concept.

10/10/2017 Q Validation Workshop 2017 12 / 27

Intuitive factoringLikert scale [Fruit] 1-2-3-4-5 [Motor Vehicle] →

correlation matrix

Items 1 2 3 4 5 6

1. Orange 1.00

2. Durian .67 1.00

3. Banana .70 .81 1.00

4. Motorcycle .11 .08 .05 1.00

5. Bus .08 .12 .09 .75 1.00

6. Car .18 .12 .22 .89 .83 1.00

10/10/2017 Q Validation Workshop 2017 13 / 27

Intuitive FactoringFactors

Items Fruit Motor vehicle

1. Orange X -

2. Durian X -

3. Banana X -

4. Motorcycle - X

5. Bus - X

6. Car - X

Correlated items → Group.More items? Impossible.

FA → objective factoring.

10/10/2017 Q Validation Workshop 2017 14 / 27

FA

• Multivariate analysis – >1 outcomes/DVs/Items.• Numerical items, e.g. Likert scale, VAS scores, laboratory

results etc.• Cluster correlated items → Factor.• Factors extracted from items → Latent (unobserved) IVs.• RQs:– Number of factors?– Strength of Item-Factor correlation (factor loading)?

• Recall – MLR: 1 DV many IVs (observed).

10/10/2017 Q Validation Workshop 2017 15 / 27

Classification

1.Exploratory FA/EFA2. Confirmatory FA/CFA

10/10/2017 Q Validation Workshop 2017 16 / 27

EFA• Exploratory analysis.• Objectives: Explore & factor items, generate theory.• Models:

–Full component model.–Common factor model.

10/10/2017 Q Validation Workshop 2017 17 / 27

EFA• Full component model

– Extraction method: Principal component analysis (PCA)– Data reduction → For other analysis.– All variances → Components.– Error NOT considered.– NOT the real FA!

10/10/2017 Q Validation Workshop 2017 18 / 27

EFA• Common factor model

– Extraction methods: • Classical: Principal axis analysis.• Other variants: Image analysis, alpha analysis, maximum

likelihood (ML).– Common + Error variances.– The 'Real' FA.– Main results:• Number of factors extracted.• Factor loadings.• Factor-factor correlations.

10/10/2017 Q Validation Workshop 2017 19 / 27

EFA• To further simplify EFA results → Factor Rotation:

– Types:• Orthogonal method – uncorrelated factors.–Varimax, Quartimax, Equamax

• Oblique method – correlated factors.–Promax, Direct Oblimin

• Obtain clear factors & factor loadings.

10/10/2017 Q Validation Workshop 2017 20 / 27

Classification

1. Exploratory FA/EFA2.Confirmatory FA/CFA

10/10/2017 Q Validation Workshop 2017 21 / 27

CFA• Confirmatory analysis.• Common factor model.• Structural Equation Modeling (SEM) analysis:–Measurement model (CFA)– Structural model (path analysis)

• Commonly ML estimation.• Model fit assessment.

10/10/2017 Q Validation Workshop 2017 22 / 27

CFA

• For example, correlation between these items:

I love fast foodI hate vegetableI hate eating fruitsI hate exercise

Obesity

Strong theoretical basis from EFA, theory, LR.

10/10/2017 Q Validation Workshop 2017 23 / 27

CFAI love catI hate snakeI love travellingI love snorkelingI support ABC football teamI love driving carI love computer gameI like to have everything normally distributedI am a strong believer of central limit theoremMy favourite food is nasi ayamI enjoy eating pisang gorengI spend most of my time in front of computerI love R

?

?

?Correlation between items? No idea → EFA

10/10/2017 Q Validation Workshop 2017 24 / 27

EFA vs CFAEFA CFA

Exploratory Confirmatory

No need theory Theory

Explore to get theory Confirm theory

Item not fixed to factor Item fixed to factor

Rotation No rotation

No Hx testing Hx testing & model fit

10/10/2017 Q Validation Workshop 2017 25 / 27

5. Reliability• Part of validity evidence.• Types:

1.Test-retest reliability2.Parallel-forms reliability3.Interrater reliability4.Internal consistency reliability

10/10/2017 Q Validation Workshop 2017 26 / 27

Internal consistency reliability• Consistent responses in a construct.• Homogenous → ↑Reliability.• Heterogenous → ↓Reliability.• Advantage: Measure 1x only.• EFA: Cronbach's alpha coefficient.• CFA: Raykov’s rho• Not reliable 0 → 1 Perfectly reliable.• Aim > 0.7.

10/10/2017 Q Validation Workshop 2017 27 / 27

Thank You